Inserting an SVD meta-layer into neural networks is prone to make the covariance ill-conditioned, which could harm the model in the training stability and generalization abilities. In this paper, we systematically study how to improve the covariance conditioning by enforcing orthogonality to the Pre-SVD layer. Existing orthogonal treatments on the weights are first investigated. However, these techniques can improve the conditioning but would hurt the performance. To avoid such a side effect, we propose the Nearest Orthogonal Gradient (NOG) and Optimal Learning Rate (OLR). The effectiveness of our methods is validated in two applications: decorrelated Batch Normalization (BN) and Global Covariance Pooling (GCP). Extensive experiments on visual recognition demonstrate that our methods can simultaneously improve the covariance conditioning and generalization. Moreover, the combinations with orthogonal weight can further boost the performances.

Improving Covariance Conditioning of the SVD Meta-layer by Orthogonality / Song, Yue; Sebe, Nicu; Wang, Wei. - 13684:(2022), pp. 356-372. (Intervento presentato al convegno European Conference on Computer Vision (ECCV) 2022 17th European Conference tenutosi a Tel AvivI, Israel nel 23–27 October 2022) [10.1007/978-3-031-20053-3_21].

Improving Covariance Conditioning of the SVD Meta-layer by Orthogonality

Song, Yue
;
Sebe, Nicu;Wang, Wei
2022-01-01

Abstract

Inserting an SVD meta-layer into neural networks is prone to make the covariance ill-conditioned, which could harm the model in the training stability and generalization abilities. In this paper, we systematically study how to improve the covariance conditioning by enforcing orthogonality to the Pre-SVD layer. Existing orthogonal treatments on the weights are first investigated. However, these techniques can improve the conditioning but would hurt the performance. To avoid such a side effect, we propose the Nearest Orthogonal Gradient (NOG) and Optimal Learning Rate (OLR). The effectiveness of our methods is validated in two applications: decorrelated Batch Normalization (BN) and Global Covariance Pooling (GCP). Extensive experiments on visual recognition demonstrate that our methods can simultaneously improve the covariance conditioning and generalization. Moreover, the combinations with orthogonal weight can further boost the performances.
2022
European Conference on Computer Vision (ECCV)
Cham
Springer
978-3-031-20052-6
978-3-031-20053-3
Song, Yue; Sebe, Nicu; Wang, Wei
Improving Covariance Conditioning of the SVD Meta-layer by Orthogonality / Song, Yue; Sebe, Nicu; Wang, Wei. - 13684:(2022), pp. 356-372. (Intervento presentato al convegno European Conference on Computer Vision (ECCV) 2022 17th European Conference tenutosi a Tel AvivI, Israel nel 23–27 October 2022) [10.1007/978-3-031-20053-3_21].
File in questo prodotto:
File Dimensione Formato  
136840352.pdf

accesso aperto

Tipologia: Post-print referato (Refereed author’s manuscript)
Licenza: Tutti i diritti riservati (All rights reserved)
Dimensione 6.68 MB
Formato Adobe PDF
6.68 MB Adobe PDF Visualizza/Apri

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/11572/361309
Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus 1
  • ???jsp.display-item.citation.isi??? 1
social impact